Aesthetics & Scales

Erik Fredner

2024-08-24

Aesthetics: so what?

  • Aesthetics (such as color, size, shape, etc.) determine how data points are visually distinguished in a plot.
    • Choosing the right aesthetics ensures that the visualization communicates the correct message.
    • For example, it would be confusing to color Democrats red and Republicans blue in a political visualization.

Scales: so what?

  • Scales control how data is mapped onto visual dimensions like the x- and y-axes.
    • This affects how easily readers can interpret the visualization.
    • Proper scaling can prevent misleading representations.

Aesthetics & Scales with Pokémon

This shows the Pokemon with the highest defense and hp in the top-right corner of the plot.

Code 
pokemon <- read_csv("../data/pokemon.csv")

pokemon |>
  ggplot() +
  geom_point(aes(x = defense, y = hp))

Modifying scales

Let’s suppose we wanted to flip that and see the Pokemon with the highest defense and lowest hp in the top-right corner.

Code 
pokemon |>
  ggplot() +
  geom_point(aes(x = defense, y = hp)) +
  # reverse the y-axis
  scale_y_reverse()

Combining scales and aesthetics

Let’s find out the name of the Pokemon with low hp and high defense:

Code 
pokemon |>
  ggplot() +
  geom_point(aes(x = defense, y = hp)) +
  scale_y_reverse() +
  # new:
  geom_text(aes(x = defense, y = hp, label = name))

Limiting scales

Code 
pokemon |>
  ggplot() +
  geom_point(aes(x = defense, y = hp)) +
  scale_y_reverse() +
  # repel the text labels:
  geom_text_repel(aes(x = defense, y = hp, label = name)) +
  # limit the x-axis to `defense` of 150 or more:
  # `NA` ("Not Available") is a missing value indicator.
  # We use it here to say that there is no upper limit on the x-axis.
  scale_x_continuous(limits = c(150, NA))

Increasing n.breaks

Code 
pokemon |>
  ggplot() +
  geom_point(aes(x = defense, y = hp)) +
  scale_y_reverse() +
  geom_text_repel(aes(x = defense, y = hp, label = name)) +
  # make it easier to identify the precise values of `defense`:
  scale_x_continuous(limits = c(150, NA), n.breaks = 30)

Color

  • We can use color to see patterns in the data by factors.
  • Let’s see if there are any patterns in the type_1 of the Pokemon in relation to their defense and hp.
  • And we’re going to filter for first generation Pokemon for simplicity.

Color by type_1

Code 
pokemon |>
  filter(generation == 1) |>
  ggplot() +
  geom_point(aes(x = defense, y = hp, color = type_1)) +
  geom_text_repel(aes(x = defense, y = hp, label = name))

Custom color

Let’s use colors associated with 🔥, 🍃, and 💧 Pokemon:

Code 
pokemon |>
  filter(generation == 1) |>
  # filter for just a few types
  filter(type_1 %in% c("Water", "Fire", "Grass")) |>
  ggplot() +
  geom_point(aes(x = defense, y = hp, color = type_1)) +
  geom_text_repel(aes(x = defense, y = hp, label = name)) +
  # use the `type_1` colors instead of the default:
  scale_color_manual(values = c(
    Water = "blue",
    Fire = "red",
    Grass = "green"
  ))

scale_color

stat_total roughly represents how powerful a Pokemon is, which makes it obvious how good Mewtwo is.

Code 
pokemon |>
  filter(generation == 1) |>
  ggplot() +
  # color the points by `stat_total` instead of `type1`:
  geom_point(aes(x = defense, y = hp, color = stat_total)) +
  # use the `viridis` color palette instead of the default:
  scale_color_viridis_c() +
  geom_text_repel(aes(x = defense, y = hp, label = name))

Size

We could also use size to represent stat_total, which makes it obvious how bad Magikarp is!

Code 
pokemon |>
  filter(generation == 1) |>
  # just water pokemon
  filter(type_1 == "Water") |>
  ggplot() +
  geom_point(aes(x = defense, y = hp, size = stat_total)) +
  # this scales the difference in size between points:
  scale_size_continuous(range = c(1, 10)) +
  geom_text_repel(aes(x = defense, y = hp, label = name))

Summary

  • Aesthetics determine how data points are visually distinguished, including aspects like color, size, and shape.
  • Scales control how data is mapped onto visual dimensions such as x- and y-axes. Proper scaling ensures that visualizations are easy to interpret and not misleading.
  • Manipulating both aesthetics and scales can reveal patterns and/or outliers in data.

https://fredner.org